Process Systems Engineering

Process Systems Engineering (PSE) is a multidisciplinary field that integrates principles of engineering, mathematics, and computer science to optimize the design, operation, and control of complex industrial processes. In PSE, advanced modeling and simulation techniques are employed to enhance process efficiency, reduce costs, and minimize environmental impact. By leveraging cutting-edge methodologies, PSE plays a pivotal role in the modern development of innovative solutions for the seamless integration of chemical, biochemical, and environmental processes. Nevertheless, the application of these solutions in cyclic adsorption/reaction or separation processes is currently constrained, prompting our exploration of this research area within the group. Despite these challenges, our group has achieved remarkable productivity and garnered international recognition.

Our research contributes significantly to the application of machine learning, particularly ANNs, in addressing complex cyclic adsorption processes. Our work critically examines the limitations of conventional mechanistic models for real-time applications, advocating for the adoption of ANNs as a promising alternative. Specifically, the utilization of Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) units allows us to handle sequential data and address long-term dependencies. Additionally, the research delves into physics-informed ML, employing Physics-Informed Neural Networks (PINNs) and Universal Differential Equations (UDEs) to solve complex numerical problems and integrate differential equations with data-driven models. Applied to industrial processes, this approach introduces innovative training methods for hybrid non-linear Partial Differential Equation (PDE) problems, offering new methodologies for addressing distributed-parameter challenges in chemical engineering. Indeed, these hybrid models can serve as effective surrogates for faster prediction of process performance, opening avenues for process optimization and control. Our group has contributed to new algorithms for metaheuristic techniques and has also advanced statistical treatment for the data obtained through such algorithms. These novel optimization techniques find practical applications in the chemical industry, including SMB and PSA. Another aspect of our work involves the use of these optimization tools to determine model and process parameters, aligning model predictions more closely with experimental observations. Moreover, our research has made substantial contributions to the control and optimization of these cyclic adsorption processes. A ground-breaking strategy has been developed to identify transfer functions in SMB units, enabling the application of classical linear Model Predictive Control (MPC) based on these functions. This approach not only addresses the challenges of traditional MPC or PID controllers but also introduces a modified MPC prediction strategy with a switching system, enhancing the handling of dynamic behaviors. Despite the potential superiority of non-linear model-based predictive control (NMPC) over linear MPC, NMPC's practical implementation faces hurdles due to its complexity and computational demands. Our work has introduced a nominally stabilizing MPC, or infinite horizon model predictive control (IHMPC), for SMB systems, representing a significant advancement in an area lacking prior research. Additionally, for PSA processes, a novel approach has been proposed for simultaneous control and optimization. This comprehensive approach significantly enhances the effectiveness and economic efficiency of cyclic adsorption processes, laying a robust foundation for future advancements in this dynamic field.

Major projects in this research area include: